Extension Theorems for Functions of Vanishing Mean Oscillation Citation Holden, Peter James (1987) Extension Theorems for Functions of Vanishing Mean Oscillation. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/f7k9-rh88. https://resolver.caltech.edu/CaltechTHESIS:11132015-131516478 Abstract A locally integrable function is said to be of vanishing mean oscillation ( VMO ) if its mean oscillation over cubes in R d converges to zero with the volume of the cubes. We establish necessary and sufficient conditions for a locally integrable function defined on a bounded measurable set of positive measure to be the restriction to that set of a VMO function. We consider the similar extension problem pertaining to BMO (ρ) functions; that is, those VMO functions whose mean oscillation over any cube is O (ρ(ℓ( Q ))) where ℓ( Q ) is the length of Q and ρ is a positive, non-decreasing function with ρ(0 + ) = 0. We apply these results to obtain sufficient conditions for a Blaschke sequence to be the zeros of an analytic BMO (ρ) function on the unit disc. Item Type: Thesis (Dissertation (Ph.D.)) Subject Keywords: Mathematics Degree Grantor: California Institute of Technology Division: Physics, Mathematics and Astronomy Major Option: Mathematics Thesis Availability: Public (worldwide access) Research Advisor(s): Goddard, William A., III (advisor) Wolff, Thomas H. (co-advisor) Thesis Committee: Goddard, William A., III (chair) Beauchamp, Jesse L. Grubbs, Robert H. Bercaw, John E. Wolff, Thomas H. Defense Date: 14 May 1987 Record Number: CaltechTHESIS:11132015-131516478 Persistent URL: https://resolver.caltech.edu/CaltechTHESIS:11132015-131516478 DOI: 10.7907/f7k9-rh88 Default Usage Policy: No commercial reproduction, distribution, display or performance rights in this work are provided. ID Code: 9279 Collection: CaltechTHESIS Deposited By: Benjamin Perez Deposited On: 13 Nov 2015 22:06 Last Modified: 16 Apr 2021 22:18 Thesis Files Preview PDF - Final Version See Usage Policy. 9MB Repository Staff Only: item control page

EXTENSION THEOREMS FOR FUNCTIONS OF VANISHING MEAN OSCILLATION Thesis by Peter J. Holden In Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy California Institute of Technology Pasadena, California 1987 (Submitted May 14 1987 ) '--- 11 - Acknowledgements I am very grateful to my thesis advisor, Prof. Tom Wolff, for his help, encouragement and patience and for giving me enthusiasm for this subject. I would also like to thank Prof. W .A.J . Luxemburg for his encouragement and for his enjoyable classes in functional analysis. - 111 - Extension theorems for functions of vanishing mean oscillation Peter J. Holden Department of Mathematics California Institute of Technology Abstract A locally integrable function is said to be of vanishing mean oscillation (V MO) if its mean oscillation over cubes in Rd converges to zero with the volume of the cubes. We establish necessary and sufficient conditions for a locally integrable function defined on a bounded measurable set of positive measure to be the restriction to that set of a V MO function. We consider the similar extension problem pertaining to BMO(p) functions; that is, thos e V MO functions whose mean oscillation over any cube is O(p(L(Q))) where £( Q) is the length of Q and p is a positive, non-decreasing function with p(o+ ) = o. We apply these results to obtain sufficient conditions for a Blaschke sequence to be the zeros of an analytic BMO(p) function on the unit disc. - lV - Contents Page Acknowledgements (ii) Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (iii) 1. Introduction ............................. . .............. _ . .... ..... ...... 1 2. Preliminary results 5 3. Proof of theorem I . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 4. Extension to BMO(p) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 5. The zeros of analytic BMO(p) functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 - 11. Introduction Let F be a locally integrable function on Rd and let Q be a cube in Rd with sides parallel to the axes. (We denote the set of all such cubes in Rd by CS'.) We denote the Lebesgue measure of Q by IQI and the length of Q by f(Q). We denote the average of F on Q by FQ ; that is FQ = ibi JQ F dt. We say F is of bounded mean oscillation (abbreviated EMO(R d) or simply EMO) if 1 r sup -1 1 IF - F Q I Q E ~' Q JQ We denote this supremum by ll F!I* . (1.1) < 00. II II* defines a norm on EMO and EMO is a Banach space with respect to this norm. (We identify functions which differ by a constant.) If in (1.1) we restrict the cubes to be dyadic we obtain the space dyadic-EMO. (By a dyadic cube we mean a cube of the form Q = {kj < Xj < (kj + 1)2-n; 1 ::; j ::; d} where n and kj, 1 ::; j ::; d, are integers .) The function space EMO was introduced in 1961 by John and Nirenberg [7] who proved the following fundamental theorem: Theorem 1.1 Let F be a locally integrable function on Rd , and for each n E Z deflne: P:n (F) = inf { aER Then, (1) F E EMO if (2) sup P:n (F) < oo . nEZ J 1 ~,\ : e(Q)sup inf - e>.IF - al ::; 2- n IQI Q and only if, < 2} - 2The implication (2)=>(1) is straightforward while (1)=>(2) is obtained by means of a Calderon-Zygmund stopping time argument. (This result and other basic results on EMO can be found in [4] and [12].) A closed subspace of EMO that we will be mainly concerned with, is the space of functions of vanishing mean oscillation (V MO) which was introduced by Sarason in [11] and is defined as: . VMO ={F E EMO: hm( sup - 1 0--+0 QE'Ji' Q 1 IF - FQ I) - 1 1 Q = O} l (Q) <8 Equivalently, by the theorem of John and Nirenberg, F E V MO if and only if FE EMO and limn--+oo fin (F) = 0. A bounded function p : R + --+ R + is called a growth function if p is nondecreasing and satisfies p(o+ ) = 0. Furthermore, we will always assume p(2t) ::; 2p(t).We define EMO(p)={F E VMO: sup I I t( Q E ~' Q p £ Q We define p(t) = t ))f IF -FQ I < oo} Q J/ P~~) d8 and say pis regular if 3 C > 0 such that p(t) ::; Cp(t). If E is a Lebesgue measurable subset of Rd of positive measure (throughout we will always assume E has positive measure unless stated otherwise), we can ask for necessary and sufficient conditions for a locally integrable function defined on E to be the restriction to E of a function in EM 0 (Rd). This characterization was given by Wolff [14] and is based upon a technique due to Rubio de Francia [10] which generalizes Jones' factorization theorem for Ap -weights [8]. The main result of this dissertation is to obtain a similar characterization for V MO functions and this is the content of the following theorem: -3 Theorem I Let E be a bounded measurable subset of Rd and let f be a locally integrable function defined on E. For each n E Z defi.ne: µn(f) =inf { _!_ : ). sup 1 inf - - / £(Q)s;2-n aER JQJ e>- IJ-al < 2} QnE Then the following are equivalent: ( 1) f is the restriction of a V M 0 function on Rd to E (2) supnEZ µn(f) < oo and Iimn _, 00 µn(f) = 0 The proof of this theorem consists of two parts. In the first part we obtain an extension to a function F which is a V MO function relative to a certain net of cubes (in particular, F will be in dyadic VMO(Rd)). We then obtain an extension for each translation of this net of cubes and the second part of the proof consists of averaging these extensions to obtain an extension to VMO(Rd). We also obtain a similar characterization for functions in BM 0 (p): Theorem II Let E be a bounded measurable subset of Rd. Let f be a locally integrable function defined on E and define µn (!) as in Theorem I. If p is a growth function satisfying : (2) inf t >O p(t) Jlog ti > 0 (3) ::J ). > 1 such that for all m E Z and for all cubes Q,f(Q) < 2 - rn with - 40 < \Q n El < IQ!/>. we have inf p(2 - n) Ilog \{x E Q n E: n>m sup x E Q' \Q'I >>.}\I>p(2 - m)llog IQ n E\I \Q' n E\ IQI t ( Q ') :::;2- n then f is the restriction to E of a function in BMO(P). In particular, if p 1s regular then f is the restriction to E of a function in BMO(p). Finally we examine some uniqueness prop erties of these extensions and consider various applications of the above results to Blaschke sequences and the zero sets of analytic BMO(p) functions on the unit disc . Throughout C will denote a positive constant which will be independent of the variables in the equation in which it occurs but which may be different at each occurrence. -52. Preliminary Results Let E be a measurable subset of Rd and let S' be a collection of cubes in Rd with Definition: (1) If Fis a locally integrable function on Rd, we define the maximal function of F relative to S' by J (M~F)(x) = sup - 1 X EQ 1 QE~ Q 1 Q F dt for all x E Rd If S' = S'' , this is the usual Hardy-Littlewood maximal function. (2) If f is a locally integrable function on E, we define the maximal function of f relative to S' by (m~J)(x) = sup XEQ QE~ 1 J f 1 Q 1 QnE dt for all x E E. Definition: (1) Let w be a positive locally integrable function on E and let 1 < p < oo. We say w is an Ap (E)-weight relative to S' if < ()() (2.1) and we denote the collection of all such weights by Ap(E, S<). If E = Rd and S' = S'' we abbreviate Ap(E, S') by Ap and say w is an Ap-weight. (2) We say a positive locally integrable function w is an A 1 (E)-weight relative to S' if -6 We denote the collection of all such weights by A1(E,8'). We record some properties of Ap (E, 8')-weights in the following proposition Proposition 2.1 1 (i) If w E Ap(E,8') then w E Ar(E,8') for all r > p and (tJ-)v=T E A __E__ . p-1 (iii) If w E Ap then F = log w E EMO. By the theorem of John and Nirenberg (Theorem 1.1), if F E EMO there exists {) > 0 such that e 8 F E Ap. (iv) We mention here the following result of Coifmann and Rochberg [3] : If F E Li0 c(Rd) and Mf(x) < oo a.e., then for each 0 < 8 < 1, (Mf) 8 E A 1 . (We prove a similar result in lemma 2.1 below ) . Definition: Let 1 < p < oo and let w E Ap(E, 8'). We say w satisfies a reverse Holder inequahty if there exists f. > 0 such that wl+c E Ap(E, 8'). Remark : If w E Ap then w satisfies a reverse Holder inequality with f. depending on p and the supremum in (2.1). This fact may be deduced from (2.1) by a repeated application of a Calderon-Zygmund stopping time argument. See [1], [9]. The next theorem is a variation of a theorem of Muckenhoupt [9]. The proof is the same and so will be omitted. Theorem 2.1 Let 1 < p < oo and let w E Ap (E, 8'). If w satisfies a reverse Holder inequality then - 7there exists a constant C > 0 such that (2.2) and (2.3) p where q = - p- 1 By a theorem of Rubio de Francia [10] ,(2.2) and (2.3) imply that there exist w 1 , w 2 E A 1 (E,8') such that w = w1w~ -p · We s ummarize what we need from the above in the following corollary Corollary 2.1 If w E Az (E, 8') and w satisfies a reverse Hold er ineq uality th en there exist w 1 , w 2 E Az(E, 8') such that w = ~ · We are now in a position to give the EMO ex.tension theorem of Wolff [14] . Theorem 2.2 If f is measurable on E, then the following are equivalent: ( 1) f is the restriction of a BM 0 function on Rd to E (2) 3..\ > 0 such that sup _ l_ f Q E~' IQI JQnE /lf - 1CJAE 1 f QnE f l < CXl (3) 3 ..\ > 0 such that sup inf - 1 QE~' aE R 1 IQJ QnE e,\lf-a l <(X) -8We give the proof as it provides one of the basic steps needed m provmg theorem I. Proof The equivalence of (2) and (3) are straightforward and the implication (1)=> (3) is similar to the proof of (1) => (2) in theorem I which we give in §3. (2)=> (1): (2) implies that w = e¥ E A 2 (E, 8- 0 ) and satisfies a reverse Holder inequality. Hence by corollary 2.1, there exist w1,w2 E A 1 (E,8'o) such that w = e¥ = ~- Define Wi = M ;s1(xE wi) t, i = 1, 2 . By Proposition 2.1 (iv), W 1, W2 E m;s1(wi), a.e. on E, (i = 1,2), it follows that 3 g E L 00 (Rd),g > 0 such that = e~ '.:? Define F = a.e. on E. "f {log g + 2 log (Wi/W2)}. Then F = f a.e on E and by Proposi- tion 2.1 (ii) and (iii), F E BMO(R d) . • Finally we prove 2 lemmas which are needed in the next section. The first is a variation of the theorem of Coifmann and Rochberg mentioned above whi le the second is based on lemma 2. 2 in [5]. For each k E Z we define Dk = {Q E 8'': Q dyadic, £(Q) = 2 - k}. Lemma 2.1 Let m, n E N with m > n and let S' = {Q : Q = LJ Q;E D,,, Qi and if£( Q) :::::; 2 - n then Q ~ Qn E Dn} - 9Then (1) Given Q E S', ::JC > 0 and Q1 E S' such that Q ~ Q1, IQ1 I :S CIQI and whenever Q2 E S' satisfies IQ2 n QI > 0 and IQ2 n Q~ I > 0 there exist Q3 E S' such that Q, Q2 ~ Q3 and jQ3I :S CIQ2I- (The constant C depends only on the dimension.) (2) \:/8, 0 < 8 < 1, 3 C s > 0, depending only on 8, such that Proof (1) If t'.(Q) 2 2 - n then we take Q 1 to be that cube with the same center as Q and of length 3£( Q) . If t'.( Q) < 2 - n then 3Qn E Dn with Q ~ Qn. If dist( Q, oQn) 2 t'.( Q) we can again take Q1 as above. In the remaining case it is not hard to see that there exists cubes {Qi} in S' satisfying IQ n Qil > 0 and !£(Q) :S t'.(Qi) :S 4t'.(Q) and such that if we take Q1 to be the completion of the Qi to a cube in S' then Q1 satisfies (1). (2) To prove (2) it suffices to show ::J C > 0 such that VQ E S', Fix Q E S' and let Q1 be as in (1) and let 91 = 9XQ 1 , 92 = (1 - xQ 1 ) 9 s o that 9 = 91 + 92· Claim: ibi JQ (M~9i) s dt :S C s infx EQ (M~9)s, i = 1, 2. Proof : i = 1: The weak-type estimate for the Hardy-Littlewood maximal function implies I{ x E Q1: (M~91) s > >-} I :S C\Q1\ ( \ l 6 0 ) - 10 - where C depends only on the dimension and Ao = ( ! J JQ g1 dt) I 1 1 (M~g1) 8 dt :::; >..o lQ1 I+ c >. t IQ1 I Qi ~ c, c~. 1 1Q. g, dt :::; c 5 (M~ gi) 8 (x) 1 8 . This implies r= j>..o r >.. - t d>.. Vx E Q1 i = 2: Fix x E int(Q) .Then (1) of the lemma implies that whenever Q 2 E 8' contains This implies :::; C inf (Mc.sg) (y) ! y EQ 1 8 . 8 IQ[ Q (Mc.s g2) dt :::; C ;~~ (Mc.s g) (y) and this proves the claim in the case i = 2. (2) of the lemma now follows from the claim and the fact Lemma 2.2 Let E be a measurable subset of the unit cube Qo with 0 < IE I < l. Then if o < {3 < log l / IE I, 3H E V MO(Qo), l Hll * ::; C o s uch that : (1) 0 :::; H :::; {3, supp(H) ~ Q, H = {3 on E (2) supq, e(Q) 2 1 l~I f q H dt :::; C o. Proof W.l.o.g we may assume IE I :::; 2 - 4 d (otherwise we may take H to be constant ). Let {Qi} be the maximal subcubes of Q for which IQ i n E l> ! /Qi /· For - 11 each j ;:::: 1 choose nj so that L IQil[ > 2i[log IU Qil [ [log i?_nj and define di > = {Qi: nj :S i< nj + i} so that LQEGU> IQI :S 4- ,BjdlQoJ where {Jj = 2i fJo , fJo = [21d [log I U Qi I [] and [ ] denotes the greatest integer function. For each j we now construct a sequence of generations { (j ) (2) Suppose G i has been defined. For each Q E G/(") },Bji= l as follows: G/ let Q(k) denote that ( ") dyadic cube of length 2k£(Q) containing Q. Choose k minimal so that (j) . . We define Gi + l to be the maximal cubes m 4- d { (k) Qr (j)} : Qr E Gi . We note that L {JQI: Q EG~J)-i- 1} :S L { JQ I : Q EG~;_, } :S 2 - d Now fix i, 1 :S i :S {Jj. Let G~i> L {JQI : Q E G~:- i- 1} = {Qk}~= l and we assume these cubes are indexed so that IQrl ;:::: JQ s l whenever r < s. Let rk,i E C 00 satisfy (i) 0 :S rk,i :S 1, ~ ~ (ii) rk,i = 1 on Qk,supp(rk,i) s;;; Qk where Qk denotes that cube with the same center as Q k and of length 3£( Q k). ork · (iii) \ ~ \:S C / £(Qk), uxz V l :S l :S d. - 12 - Now define Al ,i = r1 ,i and define b1,i = r1,i bk i = Tk i (1 - Ak - 1 i) ' ' ' 2 < k < N. U{ Qk: Qk E G (j) } Define aj = ""!3 L...i ~. l "" L., QkE c'.il bk,i and note that aj = {3j on 1 We now define H = min(~= j ?'. 1 a· 2 ~ , f3o). • Remark : Let min{£( Qi) : Qi E G~;J} = 2-n and let Q be a cube with£( Q) = 2 - m. Then for all m > n , Proof: For any xo E Q, Now I a:k ,i I :::; c / £(Qk) uxz -this follows from the definition of the bk,i and the fact I dAk,i I :::; C / R(Qk) which can be established by .induction. Furthermore there are dxz at most a fixed number of cubes in any G~i) which intersect Q. If Qk, is any such cube and Qk 2 is a generation cube containing Qk, then for all x E Q n Qk 2 and hence - 13 - 3. Proof of Theorem I Theorem I Let E be a bounded measurable subset of Rd and let f be a locally integrable function defined on E. For each n E Z define: µn(f) = inf { _!__: ,\ inf __!_ sup f(Q)9 - n aER r IQI J QnE < e.\lf - al 2} Then the following are equivalent: (1) f is the restriction of a V M 0 function on Rd to E (2) supnEZ µn(f) < 00 and limn __, 00 µn(f) = 0 Proof Without loss of generality we will assume E is contained in the unit cube (1) =? (2): Let F E V MO with FxE = J and for each n E Z define 1- ( < Tin(F) = inf { _!__ : sup inf e.\IF -al A f(Q)::; 2 -n aER IQI } Q < 1 llFl l*,n = sup f(Q)::;2-" -1 1 Q 2} 2} JrQ IF- Fq l Since FE EMO, :JC > 0 such that \In ~ 0, ll Fll *, n::; C and limn __,oo ll Fl l* ,n = 0. By Theorem (1.1), 3C1 > 0 such that whenever 0 < >.. < Ci/ llFl l,, ,n we have sup _1_ e(Q):s: 2- .. IQ I r e.\ IF - Fq I < 2 }q Hence Tin* (F) ::; llF ll *,n/ C1. Since Tin (F) ::; Tin * (F) and µn (!) ::; Tin (F), it follows that µn (!) ::; C for n = 0, 1, 2,... and limn __, 00 µn (!) = 0 and this proves (1) =? (2). - 14 - Proof of (2) ==> (1): Part (i): Extension to dyadic-VMO. Let p be a bounded growth function satisfying p(2t) :S 2p(t), \::It > 0 and µn(f) :S p(2-n), \::In 2". 0. Then (2) implies there exists a sequence p,n}n>O' 0 < >-n I oo such that sup and _l_ l(Q):=;2 - n r IQI JQ n E e.A ,,l f - /qnE I <2 Define a sequence nk ~ N by the condition p(z - n) :S 2 - k if and only if n 2". nk. To simplify the notation we will write Ak for Ank . Now define S'o = { Q : Q = LJ{ Qi : Qi E Do}} and for each k 2". 1 , S'k = { Q: Q = LJ{Qi: Qi E Dnk +i} and if £(Q) :S 2 -n, then 3 Qk E Dnk s.t . Q ~ Qk} For each n = 0, 1, 2, ... we define f n = LQ E D n f QnE XQ nE . Lemma 3.1 There exists C > 0, depending only on the dimension, such that for all k 2". 0, (1) SUP Q E"h+ 1 ( - 1 - r IQ \ JQnE Proof Fix j E N eAk(fnk+l - j ,,k) ) r (~ e - Ak(j,.k+ I - j,.k) ) < c \Q I J Q n E - 15 Claim: JC ·> 0 such that for all .A ::::; Af and for all Q, l'.(Q) 2: 2 - ni, 1 IQI -1 e >.(!-Jn 1. ) 1 IQI - QnE 1 e ->.(!-Jn 1. ) QnE <C (3.1) - This implies This implies (3.1) since Now V.A S Af and VQ, l'.(Q) S 2 - n-;, Q ~ QJ E Dn, we are given _ 1_ r eA(f-fqnE) r _l_ IQI J QnE IQI JQ nE e - A(f-fQnE) < C - and hence ( -1 IQI 1 e >.j) ( -1 1 IQI QnE Qn E e - >.j) < C (3.2) - Note that if;' = 0, (3.2) holds for all Q and for all .A ::::; .A o. Now (3.2) implies that for all Q ~ Qi E Dn1 and for all >.. :S Af -1 1 IQI QnE Since e >. (! - fn 1. ) - 1 1 IQI Q n E e - >. (! - j,,.) ' < C (3.3) - 16 - (and similarly for -I 1 I { e - >..Unk + 1 - ink)) we see that (1) follows from (3 .1) and Q lQnE (3.3). Now (3.1) and (3.2) imply that V>..:::; >..o, \IQ with t(Q) ~ 1, (IQI11 _ e>..fo Qn E < _ )(11 IQI _ e->..fa QnE 1 1 - ( IQI Qn E e>..U - fo) - )( IQI 1 _ )< 1 e->..(f - fo) QnE )( IQI 1 )( IQI 1 _ 1 e>..f Qn E _ 1 Qn E e->..f ) < C. and this gives (2) which completes the proof of the lemma. To simplify the notation we set Mk(g) = M'<!k (g) and mk (g) = m '<! k (g) . Corollary 2.1 implies the following : For each k = 0, 1, 2 ... there exist functions Uk, Vk such that (i) uo = exp (>..of Vo 2 o) Now for each k = 0, 1, 2, ... and for each x E R d we define Uk(x) = Mk(XEUk)(x) and where Uo wo Vo = - - 17 Hence, 2 + - - log wk ).k - 1 and Uo) ~ + Ao2 log wo 4 ( Jo = -log Ao Vo a.e. on E. Now lemma 2.1 (2) implies ::JC > 0 such that and so by Proposition 2.1 (ii),(~~)~ E A2(E,8''). Then, as in Proposition 2.1 (iii), we conclude that In particular since Uk, Vk are constant on dyadic cubes of length 2 - nk, we have log (Uk/Vk) Edyadic-V MO. Claim : For each k 2:: 0 Wk is the restriction to E of a function wk where log wk E dyadic-V MO. Proof :For each x EE let Qk(x) denote the dyadic cube of length 2 - nk containing x. If IQk(x) n El > 0, then and and hence IQk(x)I I log wk(x)I ::; log C + log JQk(x) n El - 18 Now lemma 2.2 implies 3Hk(x) E VMO(Qk(x)) satisfying (i) llogwk(x) - Hk(x)I :::; C (ii) IQl I. sup £(Q)?_£(Qk) r Jh(x) dt:::; Co JQ We now define otherwise. It is easy to check Hk E dyadic- V MO with llHk II * :::; Co and I log Wk (x) - Hk(x) I :::; C, \Ix E E. This implies 3Rk(x) EL 00 (Rd) with llRk l 00 :::; Co and which is constant on dyadic cubes of length 2 - nk and satisfies Rk(x) = logwk(x) - Hk(x) a.e. on E. In particular, Rk E dyadic - V MO with llRkll * :::; Co, Vk . Since Hk is supported on finitely many cubes in Dnk the function Wk = exp(Rk + Hk) satisfies log Wk E dyadic - V MO, I log Wk ll* :::; Co and WkXE =wk a.e. and the claim now follows. Now define (3.4) I where Gk = 2log (Uk /Vk)2 and by >- -1 we mean Ao. Since Gk+Rk+Hk E dyadic V MO with I Gk+ Rk + Hk II* :::; Co and since Lk?_O Ak~i < oo, it follows that FE dyadic - V MO and llFl *:::; Co Furthermore =f Hence F is a dyadic- V M 0 extension of f. a.e. on E. - 19 - Part (ii): Extension to non-dyadic V MO. Let Q 0 denote the unit cube in R d. If a E Qo and Q is any cube we define Q (a) ( a) ={x+a:xEQ} ( •xJ = {Q Dn ( a) S'n (a) ={ Q :QES'n}, : Q E Dn}, The proof of part (i) above applied to each net of dyadic cubes D!"J establishes the following : ( <>) (o) 3 functions Gk , H k Va E Qo, Vk :?: 0, such that ( ,, ) ( 0) 1) Gk (x - a) as a function of x belongs to dyadic-VMO, II Gk II. ::; Co and (a)• 1s constant on cubes Q Gk ( (.x) .. are included in the Gk (") ( E Dnk. We can assume that the bounded functions (•.• ) ) Rk ( .) ( ,, ) (.,) E Dnk ,IQk n El>O, ,, ) 2)0neachQk (·>J = min ' ( (<>) (3). . = 2J (3 (o) L a.k _ J_ ,_ . 2J ' (<>) < where and, j ?'. l ) (3 [ - l IQk(u) I 1 2d og I Q~") n E l 1 (·» I (<> ) o an d ·mt h.1scasewesetaj,k (a) fJ(·» = 0. 3 ) If IQk n E = OthenHk = = 4) If we define F (<>) ' 1 _L_.A_ _ k ?'. O ( Gk( a) +Hk(a)) k- 1 then F«•) (x - a) E dyad ic - V MO F(u) XE = f and We now define Vk :?: 0, J:?: 1 aj,k(x) = 1 Cl'.EQo (o) aj ,k (x) da , a.e. on E. - 20 - /3j = . J (a) /3j da.' and Hk = min 0tEQo (La~·/ , f3k) J~l Lemma 3.2 Hk(x) = J ( <>) Hk (x) da. a.e. on E. 0tEQo Proof The last statement in the lemma follows from the fact that Va. E Q0 , (<>) «•l < ~ aj,_k /3k - L 2J a.e. on E J ?'. 1 and so To show Hk(x) E V MO, it suffices to show each aJ,k E V MO, ll aj,kll* ~ Co. (<') Fix k 2: 0, j 2: 1 and let E > 0. For each a. E Qo, aJ,k (x - a.), as a function of (<>) x, belongs to dyadic-V MO with l aj,k (x - a.) II ~ ,dyadic ~ Co and furthermore on ( ") (<>) ( ") ( ") each cube Qnk E Dnk , aj,k E V MO(Qnk ). Hence ::Jn°' E N such that whenever 0 (Q) (<>) an d {, _ Q nk < 2 - n.,, we h ave Q C Choose N 1 E N so that the set has measure 2: (1 - E). - 21·., - cube with 2 - (N+l )::::; f (Q) < 2-N and write ( <> ) {3j a~.~~ = L L i= l bz,i G(J' , a ) i l(Q1)~l(Q) Then 1 r (a ) (a) 1 la1 ,k (x)I ::::; log £(Q) ::::; C.N , (<>. ) IQ I JQ laj,k - a j ,k I ::::; Co and Let S1 = {a E Q o : IQ n Q~:) I < IQI, VQ ~:) E D~:)} C \Q\2nkd. Hence ::::; C E. Now !. Qo\(So U Si) ( !. 1 IQI Q (•>) (n) ' ' \a1 k - (a J k ) I dt ) Q ::::; CIQo\ (SoU Si) I::::; C E. Also d a ::::; C I So I E ~ C E and note that IS1 \ ::::; - 22 and hence and it follows then that aj,k E VMO, llaj,kll * < Co and this completes the the proof of the lemma. Lemma 3.3 Given n let Q be a cube of length ::; 2 - n and let k be such that nk ::; n < nk+l · Then for all x, y E Q (1) IG j ( x) - G j (y) I ::; Ix - y 12n,;, V 0 ::; j ::; k. (2) l~ I I Q IGk+l - (Gk + dQI ::; c (n - nk) 2(nk - n) + c (3) Vj > k + 1, l~ I IQ IGj - (Gj)Q I ::; C. We first note that lemma 3.3 implies that the function L _!!_j_ E BMO(P). j ?_ O Aj-l Indeed, given Q as in the statement of the lemma, we have from (3) that j L L _!!_j_) I::; L I _S_ - ( /Q/ Q J·>k+1 >-i - 1 >-i - 1 _ J»k+1 _ _1 Q _l /Q I J >k+1 _ j IGj - (Gj) QI Q >-1 - 1 - 23 (1) and (2) imply, k :S: Clx - YI LP (2 - ni-i) 2ni + C(n - nk)2(nk-n) p (2 - nk) + C p (2 - nk) j=O k ::::; c2-n LP (2- nj- l) (2nj - 2nj- l) + c 2nk- n(n - nk)P (2 - n) + Gp (2 -n) j=O Proof of Lemma 3.3 (1) Fix x, y E Q and for 0 ::=:; j ::=:; k, let withx,y E Q (u ) C Claim: If O'. E Aj - 1 n Aj-1 then IGj I ,,) } (a) (x) - Gj and note that (Y) I::::; C o Proof: Without loss of generality take a = 0. Recall G 1 (x) = Jog M (x J E u) 1 Mj(XEVj ), and whenever Q E CS 1 contains x, 3Q' E C:Sj containing x and y with IQ'I : : ; C IQI. This - 24 - implies c Claim : If a E Aj - t then Proof: Again we can assume a = 0. Since [x - y[ < 2 - n1 whenever Q E S'j contains x, 3Q' E S'j containing both x and y and which s atisfies From this it follows that and hence Similarly and hence Now fix j, 0 ~ j ~ k, and fix x, y E Q. If £ 2': j and a E Ai then [ G~"'l (x) ( u) Gi (y) [= O This implies [Gj(x) - Gj(Y) [ ~ 1 [ G~<> ) (x) - G~.'» (y) [ + c °'EA 1 n A_; - 1 ~ C [x - y[2n7 and t his proves (1). J c °'EA_,._, 0 [G~ ) (x ) - G~.'» (y)[ - 25 - Proof of (2): Let Bk = {a E Qo : 3Q<"'> E D~:> with Q ~ Q< •» } and note that 1 f (<>) Claim: If a E Bk then IQI Q I Gk+l - ( I Gk+l Q ~ C. ( <> ) ) 3Q' E 8'~ ~ 1 containing Q and such that JQ'I ~ CIQI. Furthermore there 0 Proof: exists aQ' E Rd such that This implies and the claim now follows. c Claim: If a E Bk then Proof : Without loss of generality we may assume a= 0. Recall, 1 1 Mk+1 (U: + 1) (x) ~ CU:+ 1(x) 1 where 1 U :+ 1 = Mk+1 (xE U :+ J (x) 0 Now if Q' E CS ~ ~ 1 is that cu be of length 2 - nk containing Q, then I < - C 2(n-nk)d Mk +1 (U2 k+l ) (x) ' I < C 2(n - nk )d U 2 (x) k+ I ' V x E Q' - 26 - This implies 1 2 and similarly for log Vk + 1 and this establishes the claim. Now J a:EQo 1 IQI f Q 1 (a ) Gk+ 1 - ( (o) G k+1 ) Q I da :S J IQI f 1 o:EBk J 1 Q i + o:E B~ IQI (a) G k+1 - f 1 ( (a ) Gk+ 1 - Q (o ) Gk+ l ( ) Q I (<x) da ) I G k + 1 Q da and this proves (2) . Proof of (3) : (o) Fix j > k + 1, a E Qo. Then :3Q1 E ';Sj and aQ 1 E R d such that This implies and hence (3) follows. This completes the proof of lemma 3.3 and theorem I. A consequence of theorem I which has useful applications is the following corol- lary : - 27 - Corollary 3.1 Let E 1 , E 2 be measurable subsets of the unit cube in Rd and suppose there exists an increasing sequence of positive numbers {An} ~=O with An -----+ oo such that for each n EN and for each cube Q with £(Q) ~ 2 - n we have Then there exists F E V MO, JJ F JJ* ~ C;.. 0 with F = 0 on Ei and F = 1 on E2. Proof Set E = E 1 U E 2 in theorem I and define if x E Ei and if log 1d~k1 2: A[tog 1/t (Q) ]+ 1 otherwise. - 28 - 4. Extension to BMO(p). 4 .1 Proof of Theorem II The first part of the proof of ( 2)::::;:.. ( 1) in theorem I establishes the existence of a dyadic-V MO extension F of f which can be written in the following form : 1 (0) (0 ) (0) where>.k ::;C p(2 - nk) andGk ,Hk E dyadic-VMO.ThefunctionsGk are (0) constant on dyadic cubes of length 2 - nk . The functions Hk lemma 2.2 from which it is clear that for each k, sup - 1 €(Q):S;t Q f - 1 were obtained from 1 I Hk(0) - (Hk(0) )Q I Q as a function of t, depends only on the geometry of the set E. The hypothesis (3) in theorem II below provides a sufficient condition for the function L Hk k ?_ O Ak - 1 constructed in the proof of theorem I to be in BMO(P) for some specified growth function p. Theorem II Let E be a bounded measurable subset of Rd. Let f be a locally integrable function denned on E and define µn (!) as in Theorem I. If p is a growth function satisfying : (2) inft >o P(t) Jlogt J > 0 (3) 3). > 1 such that for all m E Z and for all cubes Q, £( Q) < 2 - m with 0 <I Qn El< IQI/>. we have I inf p ( 2 - n) log I{ x E Q n E : n>m sup x EQ ' l(Q 1 ) ~2-n IQ'I > >.} 1\ > p(2 - m) llog IQn El I IQ' n El IQI -. 29 - then f is the restriction to E of a function in BMO(P). In particular, if p 1s regular then f is the restriction to E of a function in BM 0 (p). Proof Without loss of generality we will assume E is contained in the unit cube Q 0 in Rd. It follows from theorem 1.1 that (1) is a necessary condition for f to be the restriction to E of a function in BMO(p). We also note that (1) is a sufficient condition for the function G = L >..Gk to belong to BMO(P). k ?_O (<>) satisfy ll Hk k- l I * S C log>.. on Qk. We assume then that 0 < IQk n El < IQkl / >..and for each n > m we define sup " EQ IQ n EI and En = {x EE: 8n(x) > >..} ~(Q)~2 - n Let {Qi} be the maximal dyadic subcubes of Qk with respect to the property IQi n E l > IQi I/ >.. and note that if x E Qi n E for some i, then x E EJ for all nk S J S (log l / f(Q i )) - 1. This implies and hence - 30 As in the proof of lemma 2 .2, we can find C 00 functions {a~~~} which can be written (a) /3j as a<."' k> = 'L 1, 'L bz ,i where each bz ,i is adapted to cubes of length 2:: 2-nk+i+ i = l c'..i,a ) 1 • We then define ( <>) H< "l = 'aj,k k L 2i (4 .1) j~l (a ) If IQk n El= 0 then we define Hk (<>) = aj ,k = 0 . If IQk n El > IQkl / ..\, then we may ( .,. ) (. >) choose the a J., k to b e constant and bounded and so that Hk is given by (4.1) . We note then that in all cases there exists a constant C >.., depending only on (a) ..\,such that laj ,k I ~ C ;. . (nk+J· +1 - nk). As in the proof of theorem I , we set and H(x) = I L . \ - Hk( x ), k>O k - 1 (..\ - 1 = ..\o) It remains to show H E BMO(P) . Let Q b e any cube with 2 - n N+ 1 ~ l (Q) < 2 - nN and let N H1 = N - k-1 - 1- L Ak - 1 L k=O (a_J_,_ · k) 2J. where J= l and define H 2 = H - H 1 . Lemma 2.2 (2) implies (4.2) and so H2 E BMO(P). If S = {a E Q o : Q ~ Qk E D~~> } - 31 then by the remark after lemma 2.2, - -f. 1 IQI If a €/:. S, Q i a(.") - (a(.o)) J,k J,k Q I dt < c2nk+j+i-nN - for all a E s then Hence, 1 -IQI !. Q This implies I a (o) . J,k (o ) (a J,k . ) Q I dt < C 2 n k+ 1 + 1 - n N + C >.. (n k+ ·+ 1 - n k) IS I J - 32 Combined with (4.2) we obtain The theorem now follows from the proof of theorem I. • 4.2 Uniqueness of the BMO(p) extension Corollary 2.1 implies that the V MO extension is never unique. For BMO(p) we have the following : Theorem 4.1 Let E be a measurable subset of Rd and p a growth function satisfying limsup(p(£(Q))logl IQI IQ l--. o Qn E1 )=0 (4.3) Then whenever f E BMO(p) satisfies fxE = 0 we have f = 0 a.e. Proof : Without loss of generality we may assume f 2: 0. Suppose there exists E > 0 such that the set E 1 = {x E Q : f > E} has positive measure. For each /5 > 0, ( 4.3) implies there exists n 0 such that, For any such Q, theorem 1.1 implies l{x E Q: If - /QI > ,\}I < Co IQl2 -c,>. /p(l!(Q)) and hence l!Q l::::; Co/5 For any n, 3 Q, £(Q) ::::; 2 - n such that IQ n E1I 1 IQ I > "2 and for any such Q, we have Jf QI > ~and this contradicts (4.4) for sufficiently small 15. 2 (4.4) - 33 Remark: 1) Whenever E and p satisfy (4.2), the extension to a BMO(p) function will be linear. However we do not know if the EMO extension in theorem 2.2 or the V M 0 extension are linear. 2) Given p, it is not difficult to find a set E satisfying (4.2). In the example below we obtain E as the complement of a Cantor set which is constructed using a variable ratio of dissection. Example : • • It suffices to construct Eon the umt mterval J We can find subintervals { 1) 1; iJ} of J(o) (0) = [O, 1] in R. Fix NE N. which are of equal length and satisfy L 11;1) I= 2 - (N+l) 1 1 1 ) 2) J(") \ { U i J )}.is t h e umon . o f"mterva 1s {J(l)} . f. k sat1s ymg p (li k )I) < ( N+ 3 2 . We proceed by induction (n)} Assume { Jk (n ) have been defined. On each Jk (n + l) we remove intervals 1) length and satisfying (n) 2) Jk \ { (n+l)} U lJ is the union of intervals in (k) We define E = LJJ,k lJ { (n+l)} Jk satisfying . Let 1 be an interval and suppose Then of equal - 34 - => => log I II I I < N + n + 3 I n E - P (III) log II ~IE I ::::; P( I J~n) I) (N + n + 3) 1 < - -- - N+n+3 --t 0 as n --t oo - 35 - 5. The zeros of analytic BMO(p) functions. The following definitions and notation will be used in the sequel: H 00 (D) = bounded analytic functions on the unit disc D = {z: lzl < l}. H 00 =boundary values of functions in H 00 (D). 00 A sequence { Zj} ;': 1 ~ D is called a Blaschke sequence if L (1 - I I) < oo Zj j= l and the corresponding analytic function B(z) = rr oo bJ ( z - ~i ) j= l z·J 1-z·z J . is called a Blaschke product . For each z E D we define fz = 1 {ei·a : IO - arg zl < -(1lz l)} 2 Qz = {w: lw l;:::: lzl, E Iz} : 1 1 A positive measure µ on D is called a Carleson measure if for each z E D, A Blaschke sequence { Zj} is called an interpolating sequence if for all {>..j} E l 00 3F E H 00 (D) with F(zj) = Aj. Carleson's interpolation theorem (see [ 4] Chapter 7 ) states that { zJ} is an interpolating sequence if and only if (1) inf I Zj -=- Zk I > 0 rfk 1 - ZjZk (2) _L(l - lzi l)ozi is a Carleson measure where Oz denotes the Dirac measure at z. We will also need the following characterization of BMO(p) on the unit circle, T = {z: izl = l} (see [11]) f E BMO(p) if and only if - 36 - 1) sup lzl > l - 8 (!7r-7r 1 2 If - f(z) l 2dPz (o)) ::; C p(b) (5.1) where dP (0) = ~ 1 - lzl2 dO z 27f leie - zl2 and or equivalently 2) 3C > 0 such that Vzo E D (5.2) The purpose of this section is to establish sufficient conditions for a Blaschke sequence { zJ} to be the zeros of a function in H 00 (D) with boundary values in BMO(p) . Wolff [13] has shown that every Blaschke sequence are the zeros of a function in H 00 (D) with boundary values in V MO n L 00 and every subset of the unit circle is the zero set of a function in V MO . We note that this result follows directly from corollary 2.1 . The theorem that we will prove is the following : Theorem 5.1 Let { zk} be an interpolating sequence in D and suppose p is a growth function with p 2 regular . If 3Co > 0 such that (5.3) then 3f E EMO (p) n H 00 with f( zk ) = 0 for all Zk. - 37 We note that if Bis a Blaschke product, f E L1{T) and f(z) =fr f(O) dPz(O) then JT If B - f(z)B(z)l 2dPz (O) = l 2 If - f{z)l dPz(O) + (1 - IB(z)l 2 ) lf(z)l 2 Hence if f E BMO(p) and if ::J C > 0 such that lzl > 1 - /5 implies then Bf E BMO(p). The proof of theorem 5.3 consists of obtaining a BMO(p) function f satisfying (5.4) and so that the Blaschke product with zeros Z k is sufficiently near to 1 when lf(z) / is large. Proof of Theorem 5.1 We first show that if { zk} is a Blaschke sequence satisfying (5 .3), then there exists f E BMO(p) n H 00 satisfying (5.4) . Define a sequence {nk} ~ N by the condition p(2 - n) ~ 2-kp(l) if and only if n ~ nk. Then (5.3) implies Lemma 2.2 implies there exists 9k E C 00 satisfying and Set ~ 9k g = L..,, 22k k>O - 38 - By the remark after lemma 2.2, g E BMO(p 2 ) (see also the proof of theorem II ) and, Define f(z) = exp (- (g + ig)) where g is the conjugate function of g. (5.2) implies f E BMO(p 2 ) and !f(zJ) I ~ C 2-k whenever lzJ I > 1 - 2-nk and this establishes (5.4). We note that f E BMO(a) where a(8) = sup f(Q)~8 1 \QI f I! - fq I JQ satisfies a(t) ~ p 2 (t), a(2t) ~ 2a(t), Vt~ 0 It will be convenient to work in the upper half plane Ri though we shall retain the same notation for f and B and note that BMO(p) is conformally invariant. If Q is a cube in Ri of the form Q = {(x,y): 0 < y < a}, we define the horizontal projection of Q to be the set Q* = Q n {(x,O): x ER} and the top-half of Q to be the set 1 T(Q) = {(x,y) E Q: y > -f(Q)}. 2 If z E Ri we set Imz } Q z = { (x, y) : Ix - Re z I ~ - - , y ~ Im z 2 and Iz = Q: - 39 - Claim: If I z - ~j I < ~then lf(z) I::; Ca (l Z - Zj 6 Proof lzl). : It is clear we must have Ix - Xj I < 2yj and y < 2yj . Furthermore if Ix - Xj I < kYj then y 2: tYi . This implies which implies If (z) I ::; If (z) - h z I + lhz I ::; c a ( 1 - Iz I) + Ih . - h z I + Ih 'i I j ::; C (a (1 - lzl) + a (l - lzi l)) ::; Ca (l - lzl) Hence in this case we have (1 - Now suppose IB(z)l 2 ) lf(z)l 2 ::; Ca (l - lzl). inf I z - Zj I > ~­ j z - Zj - 6 In this case, the estimate Ilog ti ::; (1 + 21log al) (1 - t) valid for a 2 < t < 1 implies Let A n = { zj : Ix - Zj l > 2n y}, n E Z and choose N so that 2 - N < a (l Then, 2= y y~ k lz - z kl 2 = 2= y y~ c Z k E AN lz-zkl 2 + 2= 2= n> _ N+1 Zk EA n - 1 \ A n y y~ lz -z kl 2 lzl). - 40 Now since { zk} is interpolating ::; Co- (1 - lzl) Hence it remains to show 51 lf(z) I2 ::; Co- (1 Let R lzl). = {(u,v): lu - xi::; 2PIIl,O < v < 2P+ 1III} where I= Iz and where p is sufficiently large so that A~ <;;: R. Subdivide R into dyadic cubes and from the collection with one side a long the x-axis, we select those that are maximal with respect to the property of containing some Zj in their top half. We denote this collection of cubes by {Qi}. From each Qi, select a point Zj contained in T( Qi). To distinguish these points we will denote them by {wJ·}. Since {zj} is an interpolating sequence, we have for each Qi, Let Dn = { (u, 0) : 2n- 2 III ::; lu - x i < 2n-l III} and Jn = Uo::;k::;n+l Dk. Let n 1 be the smallest value of n for which there exists a Wj with Im Wj 2:: y and Furthermore n 1 - l lh - 1J,., I ::; c I: ItJ k - hk + l I k=O - 41 n1 - l ~ c I: a (IJk+1I) k=O ~ c 1~;1 Ia (III) Similarly while C II w1 Ia (III) < C IJn, Ia (III) If1"'.i I < III III and hence lf(z) I ~ C I ~;; Ia (1 - lzl) Now there exists { nk} ~ N and C > 0 such that for all Wk, Im Wk 2:: y, Hence If (z) I2 ~ Ca (1 - lzl), since f is bounded We now prove 2 lf( z)I ~ L '"kEn l1u wk :"S; y y Im wk lz - Wkl 2 ~ Ca(l - lzl ) 1 Let 8'1 = {w) : II w.J I <III' IIwj n I i::::: 2 IIwj i} and 8'2 = {wJ: II w_;I <III, II w; nII < 2 IIw3i } 1 (5.5) - 42 - Cla£m : If S' 1 i= 0, then (5.6) Proof : For each Wf E S'1, lh'" J.I :::; a(IIwi I). Theorem 1.1 implies now follows from theorem 1.1. The claim implies < :::; ca (III). Finally we consider the contribution from points in 8'2. sequence of points with the property IDm_; n I wk > ~ J lwk for some Wk E 8'2 J J Cla£m Proof : Without loss of generality we may assume ID;.t:. 1 n I wi I > ~Jlwi I where D;.t:,, = Dm, n { (x, 0) : x 2: O} Then the proof of the claim above implies Therefore, !h i:::; If v;t;, I+ lh - fJ,, , I+ Ifv;;, 1 1 fJ,,, 1 I - 43 - which establishes the claim. Hence where the second sum is taken over those Wj for which IDmk n Iw 3 I > ~ 1Iw3 I 1112 ~ c (a (II I)+ lh l) IJm,I 2 ~ Ca(III). Now (5.6) and (5. 7) imply (5.5) and this completes the proof of the theorem. (5.7) - 44 - References 1. CHRIST, M., AND FEFFERMAN, R., A note on weighted norm inequalities for the Hardy-Littlewood maximal operator, Proc. Amer. Math. Soc. 87 (1983), 447-448. 2. COIFMANN, R., JONES, P., RUBIO de FRANCIA, J., On a constructive decomposition of EMO functions and factorization of Ap weights, Proc. Amer. Math. Soc. 87 (1983) 675-676. 3. COIFMANN, R., ROCHBERG, R., Another characterization of EMO, Proc. Amer. Math. Soc . 79 (1980) 249-254. 4. GARNETT, J., Bounded Analytic Functions , Academic Press, New York, 1980. 5. GARNETT, J., JONES, P., The distance in EMO to L 00 , Ann. of Math 108 (1978) 373-393. 6. GARNETT, J., JONES, P., EMO from dyadic EMO,Pac. J. Math 99 (1982) 351-372. 7. JOHN, F., NIRENBERG, L., On functions of bounded mean oscillation, Comm. Pure and Applied Math, 14 (1961) 415-426. 8. JONES, P., Factorization of Ap weights, Ann. of Math 111 (1980) 511-530. 9. MUCKENHOUPT, B., Weighted norm inequalities for the Hardy maximal function Trans. Amer. Math. Soc. 165 (1972) 207-226. 10. RUBIO de FRANCIA, J ., Factorization and extrapolation of weights, Bull. Amer. Math. Soc. 7 (1982) 393-395. - 45 - 11. SARAS ON, D., Functions of vanishing mean oscillation, Trans. Amer. Math. Soc. 207 (1975) 391-405. 12. SARASON, D., Function theory on the unit circle, Virginia Poly. Inst. and State Univ., Blacksburg, Virginia. 13. WOLFF, T., Two algebras of bounded functions, Duke Math J. 49 321-328 . 14. WOLFF, T., Restrictions of Ap weights, preprint. (1982)